Agent Server API

Port: 48915 (internal)

The agent server handles background task execution. It communicates with the main server through ZeroMQ sockets rather than HTTP.

ZeroMQ interface

Socket	Address	Type	Direction
Session events	`tcp://127.0.0.1:48961`	PUB/SUB	Main → Agent
Task results	`tcp://127.0.0.1:48962`	PUSH/PULL	Agent → Main
Analyze queue	`tcp://127.0.0.1:48963`	PUSH/PULL	Main → Agent

Analyze request:

Published when the main server detects an actionable conversation context.

Task result:

json

{
  "type": "task_result",
  "task_id": "uuid",
  "lanlan_name": "character_name",
  "result": { ... },
  "status": "completed"
}

Proactive message:

json

{
  "type": "proactive_message",
  "lanlan_name": "character_name",
  "text": "I found something interesting...",
  "source": "web_search"
}

The agent server uses three adapters for task execution:

Adapter	Module	Capabilities
MCP Client	`brain/mcp_client.py`	External tool calls via Model Context Protocol
Computer Use	`brain/computer_use.py`	Screenshot analysis, mouse/keyboard automation
Browser Use	`brain/browser_use_adapter.py`	Web browsing, form filling, content extraction

See Agent System for the detailed architecture.